SCH -- Shape Curvature Histogram v1.0

"Shape Curvature Histogram: A Shape Feature for Celiac Disease Diagnosis"

Abstract: In this work we introduce a curvature based shape feature extraction technique, which unlike others, does not necessarily depend on a closed boundary or a defined region. While the proposed feature has been developed for celiac disease diagnosis, it can potentially be utilized in other problem domains as well. To construct the proposed descriptor, first an input color channel is sub- ject to edge detection and gradient computations. Then, based on the gradient map and edge map, the local curvature is computed for each pixel as the angular difference between the maximum and minimum gra- dient angle within a certain neighborhood. Experiments show, that the feature is competitive as far as the clas- sification rate is concerned. Despite its discriminative power, a further positive aspect is the compactness of the feature vector. The code is available upon request.